Chapter 5 Multivariate Probability Distributions

advertisement
Chapter 5: Multivariate Probability Distributions
5.1
a. The sample space S gives the possible values for Y1 and Y2:
S
AA
AB
AC
BA
BB
BC
CA
CB
CC
(y1, y2) (2, 0) (1, 1) (1, 0) (1, 1) (0, 2) (1, 0) (1, 0) (0, 1) (0, 0)
Since each sample point is equally likely with probably 1/9, the joint distribution for Y1
and Y2 is given by
y1
0
1
2
0 1/9 2/9 1/9
y2 1 2/9 2/9 0
2 1/9 0
0
b. F(1, 0) = p(0, 0) + p(1, 0) = 1/9 + 2/9 = 3/9 = 1/3.
5.2
a. The sample space for the toss of three balanced coins w/ probabilities are below:
Outcome
HHH HHT HTH HTT THH THT TTH TTT
(y1, y2)
(3, 1) (3, 1) (2, 1) (1, 1) (2, 2) (1, 2) (1, 3) (0, –1)
probability 1/8
1/8
1/8
1/8
1/8
1/8
1/8
1/8
y2
y1
0
1
2
3
–1 1/8 0
0
0
1
0 1/8 2/8 1/8
2
0 1/8 1/8 0
3
0 1/8 0
0
b. F(2, 1) = p(0, –1) + p(1, 1) + p(2, 1) = 1/2.
5.3
Note that using material from Chapter 3, the joint probability function is given by
p(y1, y2) = P(Y1 = y1, Y2 = y2) =
2
 4  3 

 y   y   3 y  y 
 1  2 
1
2
9
 
 3
, where 0 ≤ y1, 0 ≤ y2, and y1 + y2 ≤ 3.
In table format, this is
y1
y2
0
1
2
3
0
0
3/84 6/84 1/84
1 4/84 24/84 12/84
0
2 12/84 18/84
0
0
3 4/84
0
0
0
93
94
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.4
a. All of the probabilities are at least 0 and sum to 1.
b. F(1, 2) = P(Y1 ≤ 1, Y2 ≤ 2) = 1. Every child in the experiment either survived or didn’t
and used either 0, 1, or 2 seatbelts.
1/ 2 1/ 3
5.5
a. P(Y1  1 / 2,Y2  1 / 3) 
  3 y dy dy
1
0
1
b. P(Y2  Y1 / 2)  
0
1
y1 / 2
 3 y dy dy
1
1
2
 .1065 .
 .5 .
0
.5
5.6
2
0
1
.5
.5
1dy1dy2   y1 y2 .5 dy2   (.5  y2 )dy2  .125.
a. P(Y1  Y2  .5)  P(Y1  .5  Y2 )  
1
0 y2 .5
0
0
1
b. P(Y1Y2  .5)  1  P(Y1Y2  .5)  1  P(Y1  .5 / Y2 )  1  
1
1
 1dy1dy2  1   (1  .5 / y2 )dy2
.5 .5 / y2
.5
= 1 – [.5 + .5ln(.5)] = .8466.
1 
5.7
a. P(Y1  1, Y2  5)    e
( y1  y2 )
0 5
 1  y1     y2

dy1 dy 2    e dy1    e dy 2   1  e 1 e 5  .00426.
0
 5


3 3 y2
b. P(Y1  Y2  3)  P(Y1  3  Y2 )  
0
e
( y1  y2 )

dy1dy2  1  4e 3  .8009.
0
1 1
5.8
a. Since the density must integrate to 1, evaluate
  ky y dy dy
1
2
1
2
 k / 4  1 , so k = 4.
0 0
y2 y1
b. F ( y1 , y2 )  P(Y1  y1 ,Y2  y2 )  4   t1t2 dt1dt2  y12 y22 , 0 ≤ y1 ≤ 1, 0 ≤ y2 ≤ 1.
0 0
c. P(Y1 ≤ 1/2, Y2 ≤ 3/4) = (1/2)2(3/4)2 = 9/64.
1 y2
5.9
a. Since the density must integrate to 1, evaluate
  k (1  y
2
)dy1dy2  k / 6  1 , so k = 6.
0 0
b. Note that since Y1 ≤ Y2, the probability must be found in two parts (drawing a picture is
useful):
1
P(Y1 ≤ 3/4, Y2 ≥ 1/2) =

1
 6(1  y2 )dy1dy2 
1/ 2 1/ 2
5.10
3/ 4 1
  6(1  y
2
)dy 2 dy1 =24/64 + 7/64 = 31/64.
1 / 2 y1
a. Geometrically, since Y1 and Y2 are distributed uniformly over the triangular region,
using the area formula for a triangle k = 1.
b. This probability can also be calculated using geometric considerations. The area of the
triangle specified by Y1 ≥ 3Y2 is 2/3, so this is the probability.
Chapter 5: Multivariate Probability Distributions
95
Instructor’s Solutions Manual
5.11
The area of the triangular region is 1, so with a uniform distribution this is the value of
the density function. Again, using geometry (drawing a picture is again useful):
29
a. P(Y1 ≤ 3/4, Y2 ≤ 3/4) = 1 – P(Y1 > 3/4) – P(Y2 > 3/4) = 1 – 12  12  14   12  14  14   32
.
b. P(Y1 – Y2 ≥ 0) = P(Y1 ≥ Y2). The region specified in this probability statement
represents 1/4 of the total region of support, so P(Y1 ≥ Y2) = 1/4.
5.12
Similar to Ex. 5.11:
a. P(Y1 ≤ 3/4, Y2 ≤ 3/4) = 1 – P(Y1 > 3/4) – P(Y2 > 3/4) = 1 –
 14   12 14 14   78 .
1 1
2 4
1/ 2 1/ 2
  2dy dy
b. P(Y1  1 / 2,Y2  1 / 2) 
1
0
1/ 2
5.13
a. F (1 / 2, 1 / 2) 
1/ 2
  30 y y
1
0
2
 1 / 2.
0
2
2
9
.
16
dy 2 dy1 
y1 1
b. Note that:
F (1/ 2, 2)  F (1/ 2, 1)  P(Y1  1/ 2,Y2  1)  P(Y1  1/ 2,Y2  1/ 2)  P(Y1  1/ 2,Y2  1/ 2)
So, the first probability statement is simply F (1 / 2, 1 / 2) from part a. The second
probability statement is found by
1 1 y2
4
P(Y1  1 / 2,Y2  1 / 2)    30 y1 y22 dy2 dy  .
16
1/ 2
0
9 4 13
Thus, F (1 / 2, 2)    .
16 16 16
c. P(Y1  Y2 )  1  P(Y1  Y2 )  1 
1 / 2 1 y1
  30 y y
1
0
5.14
a. Since f ( y1 , y2 )  0 , simply show
1 2  y1
 6y
2
1
2 y1
5.15
a. P(Y1  2,Y2  1)    e
 y1
11 21

 .65625.
32 32
y 2 dy 2 dy1  1 .
y1
.5 1 y1
0
dy 2 dy1  1 
y1
0
b. P(Y1  Y2  1)  P(Y2  1  Y1 )  
2
2
6y
2
1
y 2 dy 2 dy1  1 / 16 .
y1
2 2
dy 2 dy1    e  y1 dy1dy 2  e 1  2e 2 .
1 1
1 y2
 
b. P(Y1  2Y2 )    e  y1 dy1dy 2  1 / 2 .
0 2 y2
 
c. P(Y1  Y2  1)  P(Y1  Y2  1)  
e
0 y2 1
 y1
dy1dy 2  e 1 .
96
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.16
a. P(Y1  1 / 2,Y2  1 / 4) 
1
1/ 2
1/ 4
0
 (y
1
 y2 )dy1dy2 = 21/64 = .328125.
1 1 y2
b. P(Y1  Y2  1)  P(Y1  1  Y2 )  
0
(y
1
 y2 )dy1dy2  1 / 3 .
0
5.17
This can be found using integration (polar coordinates are helpful). But, note that this is
a bivariate uniform distribution over a circle of radius 1, and the probability of interest
represents 50% of the support. Thus, the probability is .50.
5.18
P(Y1  1,Y2  1)  
 

( y1  y2 ) / 2
1
8 1
ye
1 1
5.19
 

 

1
1
dy1 dy 2    14 y1e  y1 / 2 dy1    12 e  y2 / 2 dy 2   23 e 2 e 2  23 e 1
1
 1

a. The marginal probability function is given in the table below.
y1
0
1
2
p1(y1) 4/9 4/9 1/9
b. No, evaluating binomial probabilities with n = 3, p = 1/3 yields the same result.
5.20
a. The marginal probability function is given in the table below.
y2
–1 1
2
3
p2(y2) 1/8 4/8 2/8 1/8
b. P(Y1  3 | Y2  1) 
5.21
P (Y1 3,Y2 1)
P (Y2 1)
 14 // 88  1 / 4 .
a. The marginal distribution of Y1 is hypergeometric with N = 9, n = 3, and r = 4.
b. Similar to part a, the marginal distribution of Y2 is hypergeometric with N = 9, n = 3,
and r = 3. Thus,
P(Y1  1 | Y2  2) 
P (Y1 1,Y2 2 )
P (Y 22 )

 4  3  2 
 1   2   0 
   
9
 3 
 
 3  6 
 2   1 
  
9
 3 
 
 2/3.
c. Similar to part b,
P(Y3  1 | Y2  1)  P(Y1  1 | Y2  1) 
5.22
P (Y1 1,Y2 1)
P (Y 21)

 3  2  4 
     
 1  1  1 
9
 
 3
 3  6 
   
 1  2 
9
 
 3
 8 / 15 .
a. The marginal distributions for Y1 and Y2 are given in the margins of the table.
b.
P(Y2 = 0 | Y1 = 0) = .38/.76 = .5
P(Y2 = 1 | Y1 = 0) = .14/.76 = .18
P(Y2 = 2 | Y1 = 0) = .24/.76 = .32
c. The desired probability is P(Y1 = 0 | Y2 = 0) = .38/.55 = .69.
Chapter 5: Multivariate Probability Distributions
97
Instructor’s Solutions Manual
1
5.23
a. f 2 ( y 2 )   3 y1 dy1  23  23 y 22 , 0  y 2  1 .
y2
b. Defined over y2 ≤ y1 ≤ 1, with the constant y2 ≥ 0.
y1
c. First, we have f1 ( y1 )   3 y1dy2  3 y22 , 0  y1  1 . Thus,
0
f ( y2 | y1 )  1/ y1 , 0  y2  y1 . So, conditioned on Y1 = y1, we see Y2 has a uniform
distribution on the interval (0, y1). Therefore, the probability is simple:
P(Y2 > 1/2 | Y1 = 3/4) = (3/4 – 1/2)/(3/4) = 1/3.
5.24
a. f1 ( y1 )  1, 0  y1  1 , f 2 ( y2 )  1, 0  y2  1 .
b. Since both Y1 and Y2 are uniformly distributed over the interval (0, 1), the probabilities
are the same: .2
c. 0  y2  1 .
d. f ( y1 | y2 )  f ( y1 )  1, 0  y1  1
e. P(.3 < Y1 < .5 | Y2 = .3) = .2
f. P(.3 < Y2 < .5 | Y2 = .5) = .2
g. The answers are the same.
5.25
a. f1 ( y1 )  e  y1 , y1  0 , f 2 ( y2 )  e  y2 , y2  0 . These are both exponential density
functions with β = 1.
b. P(1  Y1  2.5)  P(1  Y2  2.5)  e 1  e 2.5  .2858.
c. y2 > 0.
d. f ( y1 | y2 )  f1 ( y1 )  e  y1 , y1  0 .
e. f ( y2 | y1 )  f 2 ( y2 )  e  y2 , y2  0 .
f. The answers are the same.
g. The probabilities are the same.
5.26
a. f1 ( y1 )   4 y1 y2 dy2  2 y1 , 0  y1  1; f ( y2 )  2 y2 , 0  y2  1 .
1
0
1/ 2
1
0
3/ 4
1
  4 y y dy dy
1
b. P(Y1  1 / 2 |Y2  3 / 4) 
2
 2 y dy
2
1
2
1/ 2

 2 y dy
1
1
 1/ 4 .
0
2
3/ 4
c. f ( y1 | y2 )  f1 ( y1 )  2 y1 , 0  y1  1 .
d. f ( y2 | y1 )  f 2 ( y2 )  2 y2 , 0  y2  1 .
3/ 4
e. P(Y1  3 / 4 |Y2  1 / 2)  P(Y1  3 / 4 ) 
 2 y dy
1
0
1
 9 / 16 .
98
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
1
5.27
a. f1 ( y1 )   6(1  y 2 )dy 2  3(1  y1 ) 2 , 0  y1  1;
y1
y2
f 2 ( y2 )   6(1  y2 )dy1  6 y2 (1  y2 ), 0  y2  1 .
0
1 / 2 y2
b. P(Y2  1 / 2 |Y1  3 / 4) 
  6(1  y
0
2
)dy1 dy 2
 32 / 63.
0
3/ 4
 3(1  y )
1
2
dy1
0
c. f ( y1 | y2 )  1/ y2 , 0  y1  y2  1 .
d. f ( y2 | y1 )  2(1  y2 ) /(1  y1 ) 2 , 0  y1  y2  1 .
e. From part d, f ( y2 | 1/ 2)  8(1  y2 ), 1/ 2  y2  1 . Thus, P(Y2  3 / 4 | Y1  1/ 2)  1/ 4.
5.28
Referring to Ex. 5.10:
2
a. First, find f 2 ( y 2 )   1dy1  2(1  y 2 ), 0  y 2  1 . Then, P(Y2  .5)  .25 .
2 y2
b. First find f ( y1 | y 2 ) 
1
2 (1 y2 )
, 2 y 2  y1  2. Thus, f ( y1 | .5)  1, 1  y1  2 –– the
conditional distribution is uniform on (1, 2). Therefore, P(Y1  1.5 | Y2  .5)  .5
5.29
Referring to Ex. 5.11:
a. f 2 ( y 2 ) 
1 y2
 1dy
1
 2(1  y 2 ), 0  y 2  1 . In order to find f1(y1), notice that the limits of
y2 1
integration are different for 0 ≤ y1 ≤ 1 and –1 ≤ y1 ≤ 0. For the first case:
f1 ( y1 ) 
1 y1
1dy
2
 1  y1 , for 0 ≤ y1 ≤ 1. For the second case, f1 ( y1 ) 
0
1 y1
1dy
2
 1  y1 , for
0
–1 ≤ y1 ≤ 0. This can be written as f1 ( y1 )  1  | y1 | , for –1 ≤ y1 ≤ 1.
b. The conditional distribution is f ( y 2 | y1 )  11| y1| , for 0 ≤ y1 ≤ 1 – |y1|. Thus,
3/ 4
f ( y2 | 1/ 4)  4 / 3 . Then, P(Y2  1 / 2 | Y1  1 / 4)   4 / 3dy2 = 1/3.
1/ 2
5.30
a. P(Y1  1 / 2,Y2  1 / 4) 
1 / 4 1 y2
  2dy dy
1
0
1/ 4
2
 . And, P(Y2  1 / 4) 
3
16
1/ 2
3
7

2(1  y2 )dy2  167 .
0
Thus, P(Y1  1 / 2 | Y2  1 / 4)  .
b. Note that f ( y1 | y 2 )  11y2 , 0  y1  1  y 2 . Thus, f ( y1 | 1/ 4)  4 / 3, 0  y1  3 / 4 .
3/ 4
Thus, P(Y2  1 / 2 | Y1  1 / 4) 
 4 / 3dy
1/ 2
2
= 1/3.
Chapter 5: Multivariate Probability Distributions
99
Instructor’s Solutions Manual
5.31
a. f1 ( y1 ) 
1 y1
 30 y y
1
2
2
dy 2  20 y1 (1  y1 ) 2 , 0  y1  1 .
y1 1
b. This marginal density must be constructed in two parts:
1 y2
2
2
  30 y1 y 2 dy1  15 y 2 (1  y 2 )  1  y 2  0

f 2 ( y 2 )   10 y2
.
 30 y y 2 dy  5 y 2 (1  y ) 0  y  1
1 2
1
2
2
2
 0
c. f ( y 2 | y1 )  23 y 22 (1  y1 ) 3 , for y1 – 1 ≤ y2 ≤ 1 – y1.
d. f ( y 2 | .75)  23 y 22 (.25) 3 , for –.25 ≤ y2 ≤ .25, so P(Y2 > 0 | Y1 = .75) = .5.
5.32
a. f1 ( y1 ) 
2  y1
6y
2
1
y 2 dy 2  12 y12 (1  y1 ), 0  y1  1 .
y1
b. This marginal density must be constructed in two parts:
y2

6 y12 y 2 dy1  2 y 24
0  y2  1



0
f 2 ( y 2 )  2  y 2
.
 6 y 2 y dy  2 y (2  y ) 3 1  y  2
2
2
2
 0 1 2 1
c. f ( y 2 | y1 )  12 y2 /(1  y1 ), y1  y2  2  y1 .
d. Using
11
the density found in part c, P(Y2  1.1 | Y1  .6) 
1
2
y
2
/ .4dy2  .53
.6
5.33
Refer to Ex. 5.15:
y1
a. f 1( y1 )   e
 y1
dy2  y1e
 y1

, y1  0. f 2( y 2 )   e  y1 dy1  e  y2 , y 2  0.
y2
0
 ( y1  y2 )
b. f ( y1 | y2 )  e
, y1  y2 .
c. f ( y2 | y1 )  1/ y1 , 0  y2  y1 .
d. The density functions are different.
e. The marginal and conditional probabilities can be different.
5.34
a. Given Y1 = y1, Y2 has a uniform distribution on the interval (0, y1).
b. Since f1(y1) = 1, 0 ≤ y1 ≤ 1, f (y1, y2) = f (y2 | y1)f1(y1) = 1/y1, 0 ≤ y2 ≤ y1 ≤ 1.
1
c. f 2 ( y 2 )   1 / y1 dy1   ln( y 2 ), 0  y 2  1 .
y2
5.35
With Y1 = 2, the conditional distribution of Y2 is uniform on the interval (0, 2). Thus,
P(Y2 < 1 | Y1 = 2) = .5.
100
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
1
5.36
a. f1 ( y1 )   ( y1  y 2 )dy2  y1  12 , 0 ≤ y1 ≤ 1. Similarly f 2 ( y2 )  y2  12 , 0 ≤ y2 ≤ 1.
0
1
b. First, P(Y2  )   ( y2  )  , and P(Y1  ,Y2  ) 
1
2
5
8
1
2
1
2
1
2
1/ 2
1
1
 (y
1
 y2 )dy1dy2  83 .
1/ 2 1/ 2
Thus, P(Y1  12 | Y2  12 )  53 .
1
 y
c. P(Y1  .75 | Y2  .5)  .75
1
 12 dy1
1
2
 12
 .34375.

5.37
Calculate f 2 ( y2 )   y81 e ( y1  y2 ) / 2 dy1  12 e  y2 / 2 , y2 > 0. Thus, Y2 has an exponential
0
distribution with β = 2 and P(Y2 > 2) = 1 – F(2) = e–1.
5.38
This is the identical setup as in Ex. 5.34.
a. f (y1, y2) = f (y2 | y1)f1(y1) = 1/y1, 0 ≤ y2 ≤ y1 ≤ 1.
b. Note that f (y2 | 1/2) = 1/2, 0 ≤ y2 ≤ 1/2. Thus, P(Y2 < 1/4 | Y1 = 1/2) = 1/2.
c. The probability of interest is P(Y1 > 1/2 | Y2 = 1/4). So, the necessary conditional
density is f (y1 | y2) = f (y1, y2)/f2(y2) = y1 ( 1ln y2 ) , 0 ≤ y2 ≤ y1 ≤ 1. Thus,
1

P(Y1 > 1/2 | Y2 = 1/4) =
1
y1 ln 4
dy1 = 1/2.
1/ 2
5.39
The result follows from:
P(Y1  y1 ,W  w) P(Y1  y1 ,Y1  Y2  w) P(Y1  y1 ,Y2  w  y1 )
.
P(Y1  y1 | W  w) 


P(W  w)
P(W  w)
P(W  w)
Since Y1 and Y2 are independent, this is
P(Y1  y1 ) P(Y2  w  y1 )
P(Y1  y1 | W  w) 

P(W  w)
 w  1 

=  
 y1  1   2 
y1
1 y1 e  1
y1 !

 2 w  y1 e   2
( w  y1 )!
( 1  2 ) w e
w!

1 
1 





1
2 
 ( 1   2 )
w  y1
This is the binomial distribution with n = w and p =
.
1
.
1   2

Chapter 5: Multivariate Probability Distributions
101
Instructor’s Solutions Manual
5.40
5.41
As the Ex. 5.39 above, the result follows from:
P(Y1  y1 ,W  w) P(Y1  y1 ,Y1  Y2  w) P(Y1  y1 ,Y2  w  y1 )
.
P(Y1  y1 | W  w) 


P(W  w)
P(W  w)
P(W  w)
Since Y1 and Y2 are independent, this is (all terms involving p1 and p2 drop out)
 n1  n2 
 

0  y1  n1
P(Y1  y1 ) P(Y2  w  y1 )  y1  w  y1 
.
P(Y1  y1 | W  w) 

,
0  w  y1  n2
P(W  w)
 n1  n2 


 w 
Let Y = # of defectives in a random selection of three items. Conditioned on p, we have
 3
P(Y  y | p )    p y (1  p ) 3 y , y  0, 1, 2, 3.
 y
We are given that the proportion of defectives follows a uniform distribution on (0, 1), so
the unconditional probability that Y = 2 can be found by
1
1
1
1
0
0
0
0
P(Y  2)   P(Y  2, p )dp   P(Y  2 | p ) f ( p )dp   3 p 2 (1  p ) 31 dp  3 ( p 2  p 3 )dp
= 1/4.
5.42
(Similar to Ex. 5.41) Let Y = # of defects per yard. Then,



p( y )   P(Y  y, )d   P(Y  y | ) f ()d    ye! e  d   12 
y 
0
0
y 1
, y = 0, 1, 2, … .
0
Note that this is essentially a geometric distribution (see Ex. 3.88).
5.43
Assume f ( y1 | y2 )  f1 ( y1 ). Then, f ( y1 , y2 )  f ( y1 | y2 ) f 2 ( y2 )  f1 ( y1 ) f 2 ( y2 ) so that
Y1 and Y2 are independent. Now assume that Y1 and Y2 are independent. Then, there
exists functions g and h such that f ( y1 , y2 )  g ( y1 )h( y2 ) so that
1    f ( y1 , y2 )dy1dy2   g ( y1 )dy1  h( y2 )dy2 .
Then, the marginals for Y1 and Y2 can be defined by
g ( y1 )h( y2 )
g ( y1 )
h( y 2 )
, so f 2 ( y 2 ) 
.
f1 ( y1 )  
dy2 
g
(
y
)
dy

h
(
y
)
dy
g
(
y
)
dy
h
(
y
)
dy
 1 1  2 2
 1 1
 2 2
Thus, f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) . Now it is clear that
f ( y1 | y2 )  f ( y1 , y2 ) / f 2 ( y2 )  f1 ( y1 ) f 2 ( y2 ) / f 2 ( y2 )  f1 ( y1 ) ,
provided that f 2 ( y2 ) > 0 as was to be shown.
5.44
The argument follows exactly as Ex. 5.43 with integrals replaced by sums and densities
replaced by probability mass functions.
5.45
No. Counterexample: P(Y1 = 2, Y2 = 2) = 0 ≠ P(Y1 = 2)P(Y2 = 2) = (1/9)(1/9).
5.46
No. Counterexample: P(Y1 = 3, Y2 = 1) = 1/8 ≠ P(Y1 = 3)P(Y2 = 1) = (1/8)(4/8).
102
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.47
Dependent. For example: P(Y1 = 1, Y2 = 2) ≠ P(Y1 = 1)P(Y2 = 2).
5.48
Dependent. For example: P(Y1 = 0, Y2 = 0) ≠ P(Y1 = 0)P(Y2 = 0).
y1
5.49
1
Note that f1 ( y1 )   3 y1dy2  3 y , 0  y1  1 , f 2 ( y 2 )   3 y1dy1  23 [1  y 22 ], 0  y 2  1 .
2
1
y1
0
Thus, f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) so that Y1 and Y2 are dependent.
5.50
1
1
0
0
a. Note that f1 ( y1 )   1dy2  1, 0  y1  1 and f 2 ( y2 )   1dy1  1, 0  y2  1 . Thus,
f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) so that Y1 and Y2 are independent.
b. Yes, the conditional probabilities are the same as the marginal probabilities.
5.51


0
0
a. Note that f1 ( y1 )   e ( y1  y2 ) dy2  e  y1 , y1  0 and f 2 ( y2 )   e ( y1  y2 ) dy1  e  y2 , y2  0 .
Thus, f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) so that Y1 and Y2 are independent.
b. Yes, the conditional probabilities are the same as the marginal probabilities.
5.52
Note that f ( y1 , y2 ) can be factored and the ranges of y1 and y2 do not depend on each
other so by Theorem 5.5 Y1 and Y2 are independent.
5.53
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.54
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.55
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.56
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.57
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.58
Following Ex. 5.32, it is seen that f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) so that Y1 and Y2 are
dependent.
5.59
The ranges of y1 and y2 depend on each other so Y1 and Y2 cannot be independent.
5.60
From Ex. 5.36, f1 ( y1 )  y1  12 , 0 ≤ y1 ≤ 1, and f 2 ( y2 )  y2  12 , 0 ≤ y2 ≤ 1. But,
f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 ) so Y1 and Y2 are dependent.
5.61
Note that f ( y1 , y2 ) can be factored and the ranges of y1 and y2 do not depend on each
other so by Theorem 5.5, Y1 and Y2 are independent.
Chapter 5: Multivariate Probability Distributions
103
Instructor’s Solutions Manual
5.62
Let X, Y denote the number on which person A, B flips a head on the coin, respectively.
Then, X and Y are geometric random variables and the probability that the stop on the
same number toss is:
P( X  1,Y  1)  P( X  2,Y  2)    P( X  1) P(Y  1)  P( X  2) P(Y  2)  



i 1
i 1
k 0
=  P( X  i ) P(Y  i )   p(1  p )i 1 p(1  p )i 1  p 2 [(1  p ) 2 ]k 
 y1
5.63
P(Y1  Y2 ,Y1  2Y2 )  
e
( y1  y2 )
 
dy 2 dy1 
1
6
and P(Y1  2Y2 )  
0 y1 / 2
e
( y1  y2 )
p2
.
1  (1  p ) 2
dy 2 dy1  23 . So,
0 y1 / 2
P(Y1  Y2 | Y1  2Y2 )  1/ 4.
1
5.64
1 y1 / 2
y1
P(Y1  Y2 ,Y1  2Y2 )    1dy 2 dy1  , P(Y1  2Y2 )  1  P(Y1  2Y2 )  1    1dy2 dy1  43 .
1
4
0 y1 / 2
0
0
So, P(Y1  Y2 | Y1  2Y2 )  1/ 3.

5.65
a. The marginal density for Y1 is f1 ( y1 )   [(1  (1  2e  y1 )(1  2e  y2 )]e  y1  y2 dy2
0



= e  y1   e  y2 dy 2  (1  2e  y1 ) (e  y2  2e 2 y2 )dy 2 .
0
0



= e  y1   e  y2 dy 2  (1  2e  y1 )(1  1)  e  y1 ,
0

which is the exponential density with a mean of 1.
b. By symmetry, the marginal density for Y2 is also exponential with β = 1.
c. When α = 0, then f ( y1 , y2 )  e  y1  y2  f1 ( y1 ) f 2 ( y2 ) and so Y1 and Y2 are independent.
Now, suppose Y1 and Y2 are independent. Then, E(Y1Y2) = E(Y1)E(Y2) = 1. So,

E(Y1Y2 ) 
  y y [(1  (1  2e
1
2
 y1
)(1  2e  y2 )]e  y1  y2 dy1dy2
0 0

 

 y1  y2
 y1
 y1
 y2
y
y
e
dy
dy


y
(
1

2
e
)
e
dy
)e  y2 dy 2 
 1
1
2
1     y 2 (1  2e
0 0 1 2
0
 0


=
= 1  1  12 1  12   1   / 4 . This equals 1 only if α = 0.
5.66
a. Since F2 ()  1 , F ( y1 , )  F1 ( y1 )  1 1  {1  F1 ( y1 )}{1  1}  F1 ( y1 ) .
b. Similarly, it is F2 ( y2 ) from F ( y1 , y2 )
c. If α = 0, F ( y1 , y2 )  F1 ( y1 )F2 ( y2 ) , so by Definition 5.8 they are independent.
d. If α ≠ 0, F ( y1 , y2 )  F1 ( y1 )F2 ( y2 ) , so by Definition 5.8 they are not independent.
104
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.67
5.68
P(a  Y1  b, c  Y2  d )  F (b, d )  F (b, c)  F (a, d )  F (a, c)
 F1 (b)F2 (d )  F1 (b)F2 (c)  F1 (a)F2 (d )  F1 (a)F2 (c)
 F1 (b)F2 (d )  F2 (c)  F1 (a )F2 (d )  F2 (c)
 F1 (b)  F1 (a) F2 (d )  F2 (c)
 P(a  Y1  b)  P(c  Y2  d ) .
2
Given that p1 ( y1 )   (.2) y1 (.8) 2 y1 , y1 = 0, 1, 2, and p2 ( y2 )  (.3) y2 (.7)1 y1 , y2 = 0, 1:
 y1 
2
a. p( y1 , y 2 )  p1 ( y1 ) p2 ( y 2 )   (.2) y1 (.8) 2 y1 (.3) y2 (.7)1 y1 , y1 = 0, 1, 2 and y2 = 0, 1.
 y1 
b. The probability of interest is P(Y1  Y2  1)  p(0, 0)  p(1, 0)  p(0, 1) = .864.
5.69
a. f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 )  (1 / 9)e ( y1  y2 ) / 3 , y1 > 0, y2 > 0.
1 1 y2
b. P(Y1  Y2  1)  
0
5.70
 (1/ 9)e
( y1  y2 ) / 3
dy1dy2  1  43e 1 / 3 = .0446.
0
With f ( y1 , y2 )  f1 ( y1 ) f 2 ( y2 )  1 , 0 ≤ y1 ≤ 1, 0 ≤ y2 ≤ 1,
1 / 4 y1
P(Y2 ≤ Y1 ≤ Y2 + 1/4) =
y1
  1dy dy    1dy dy
2
0
5.71
1
1
2
1
 7 / 32 .
1 / 4 y1 1 / 4
0
Assume uniform distributions for the call times over the 1–hour period. Then,
a. P(Y1  1 / 2,Y2  1 / 2)  P(Y1  1 / 2P(Y2  1 / 2)  (1 / 2)(1 / 2)  1 / 4 .
b. Note that 5 minutes = 1/12 hour. To find P(| Y1 Y2 |  1/ 12) , we must break the
region into three parts in the integration:
P(| Y1  Y2 |  1 / 12) 
1 / 12 y1 1 / 12

 1dy2 dy1 
0
0
11 / 12 y1 1 / 12

 1dy2 dy1 
1 / 12 y1 1 / 12
1

1
1dy dy
2
1
= 23/144.
11 / 12 y1 1 / 12
5.72
a. E(Y1) = 2(1/3) = 2/3.
b. V(Y1) = 2(1/3)(2/3) = 4/9
c. E(Y1 – Y2) = E(Y1) – E(Y2) = 0.
5.73
Use the mean of the hypergeometric: E(Y1) = 3(4)/9 = 4/3.
5.74
The marginal distributions for Y1 and Y2 are uniform on the interval (0, 1). And it was
found in Ex. 5.50 that Y1 and Y2 are independent. So:
a. E(Y1 – Y2) = E(Y1) – E(Y2) = 0.
b. E(Y1Y2) = E(Y1)E(Y2) = (1/2)(1/2) = 1/4.
c. E(Y12 + Y22) = E(Y12) + E(Y22) = (1/12 + 1/4) + (1/12 + 1/4) = 2/3
d. V(Y1Y2) = V(Y1)V(Y2) = (1/12)(1/12) = 1/144.
Chapter 5: Multivariate Probability Distributions
105
Instructor’s Solutions Manual
5.75
The marginal distributions for Y1 and Y2 are exponential with β = 1. And it was found in
Ex. 5.51 that Y1 and Y2 are independent. So:
a. E(Y1 + Y2) = E(Y1) + E(Y2) = 2, V(Y1 + Y2) = V(Y1) + V(Y2) = 2.


b. P(Y1  Y2  3)  P(Y1  3  Y2 )  
e
 y1  y2
dy1dy 2 =(1/2)e–3 = .0249.
0 3 y2

c. P(Y1  Y2  3)  P(Y1  Y2  3)  

e
 y1  y2
dy 2 dy1 =(1/2)e–3 = .0249.
0 3 y1
d. E(Y1 – Y2) = E(Y1) – E(Y2) = 0, V(Y1 – Y2) = V(Y1) + V(Y2) = 2.
e. They are equal.
5.76
From Ex. 5.52, we found that Y1 and Y2 are independent. So,
1
a. E (Y1 )   2 y12 dy1  2 / 3 .
0
1
b. E (Y12 )   2 y13 dy1  2 / 4 , so V (Y1 )  2 / 4  4 / 9  1/ 18 .
0
c. E(Y1 – Y2) = E(Y1) – E(Y2) = 0.
5.77
Following Ex. 5.27, the marginal densities can be used:
1
1
a. E (Y1 )   3 y1 (1  y1 ) 2 dy1  1 / 4, E (Y2 )   6 y2 (1  y2 )dy2  1 / 2 .
0
0
1
b. E (Y1 )   3 y1 (1  y1 ) 2 dy1  1 / 10, V (Y1 )  1 / 10  (1 / 4) 2  3 / 80 ,
2
2
0
1
E (Y2 )   6 y2 (1  y2 )dy2  3 / 10, V (Y2 )  3 / 10  (1 / 2) 2  1 / 20 .
2
2
0
c. E(Y1 – 3Y2) = E(Y1) – 3∙E(Y2) = 1/4 – 3/2 = –5/4.
5.78
a. The marginal distribution for Y1 is f1(y1) = y1/2, 0 ≤ y1 ≤ 2. E(Y1) = 4/3, V(Y1) = 2/9.
b. Similarly, f2(y2) = 2(1 – y2), 0 ≤ y2 ≤ 1. So, E(Y2) = 1/3, V(Y1) = 1/18.
c. E(Y1 – Y2) = E(Y1) – E(Y2) = 4/3 – 1/3 = 1.
d. V(Y1 – Y2) = E[(Y1 – Y2)2] – [E(Y1 – Y2)]2 = E(Y12) – 2E(Y1Y2) + E(Y22) – 1.
1 2
Since E(Y1Y2) =
  y y dy dy
1
2
1
2
 1 / 2 , we have that
0 2 y2
V(Y1 – Y2) = [2/9 + (4/3)2] – 1 + [1/18 + (1/3)2] – 1 = 1/6.
Using Tchebysheff’s theorem, two standard deviations about the mean is (.19, 1.81).
106
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.79
Referring to Ex. 5.16, integrating the joint density over the two regions of integration:
0 1 y1
E (Y1Y2 )  
1
1 1 y1
 y y dy dy    y y dy dy
1
2
2
1
0
1
0
2
2
1
0
0
5.80
From Ex. 5.36, f1 ( y1 )  y1  12 , 0 ≤ y1 ≤ 1, and f 2 ( y2 )  y2  12 , 0 ≤ y2 ≤ 1. Thus,
E(Y1) = 7/12 and E(Y2) = 7/12. So, E(30Y1 + 25Y2) = 30(7/12) + 25(7/12) = 32.08.
5.81
Since Y1 and Y2 are independent, E(Y2/Y1) = E(Y2)E(1/Y1). Thus, using the marginal
densities found in Ex. 5.61,

 

E(Y2/Y1) = E(Y2)E(1/Y1) = 12  y 2 e  y2 / 2 dy 2  14  e  y1 / 2 dy1   2( 12 )  1 .
0
 0

5.82
The marginal densities were found in Ex. 5.34. So,
1
E(Y1 – Y2) = E(Y1) – E(Y2) = 1/2 –   y 2 ln( y 2 )dy 2 = 1/2 – 1/4 = 1/4.
0
5.83
From Ex. 3.88 and 5.42, E(Y) = 2 – 1 = 1.
5.84
All answers use results proven for the geometric distribution and independence:
a. E(Y1) = E(Y2) = 1/p, E(Y1 – Y2) = E(Y1) – E(Y2) = 0.
b. E(Y12) = E(Y22) = (1 – p)/p2 + (1/p)2 = (2 – p)/p2. E(Y1Y2) = E(Y1)E(Y2) = 1/p2.
c. E[(Y1 – Y2)2] = E(Y12) – 2E(Y1Y2) + E(Y22) = 2(1 – p)/p2.
V(Y1 – Y2) = V(Y1) + V(Y2) = 2(1 – p)/p2.
d. Use Tchebysheff’s theorem with k = 3.
5.85
a. E(Y1) = E(Y2) = 1 (both marginal distributions are exponential with mean 1)
b. V(Y1) = V(Y2) = 1
c. E(Y1 – Y2) = E(Y1) – E(Y2) = 0.
d. E(Y1Y2) = 1 – α/4, so Cov(Y1, Y2) = – α/4.
e. V(Y1 – Y2) = V(Y1) + V(Y2) – 2Cov(Y1, Y2) = 1 + α/2. Using Tchebysheff’s theorem
with k = 2, the interval is ( 2 2   / 2 , 2 2   / 2 ) .
5.86
Using the hint and Theorem 5.9:
a. E(W) = E(Z)E( Y11 / 2 ) = 0E( Y11 / 2 ) = 0. Also, V(W) = E(W2) – [E(W)]2 = E(W2).
Now, E(W2) = E(Z2)E( Y11 ) = 1∙E( Y11 ) = E( Y11 ) = 11 2 , ν1 > 2 (using Ex. 4.82).
b. E(U) = E(Y1)E( Y21 ) =  21 2 , ν2 > 2, V(U) = E(U2) – [E(U)]2 = E(Y12)E( Y22 ) –
= 1 (1  2) (  2 2 )(1  2 4 ) –
 
1 2
 2 2
=
2 1 ( `  2 2 )
(  2 2 ) 2 (  2 4 )
, ν2 > 4.
 
1 2
 2 2
Chapter 5: Multivariate Probability Distributions
107
Instructor’s Solutions Manual
5.87
a. E(Y1 + Y2) = E(Y1) + E(Y2) = ν1 + ν2.
b. By independence, V(Y1 + Y2) = V(Y1) + V(Y2) = 2ν1 + 2ν2.
5.88
It is clear that E(Y) = E(Y1) + E(Y2) + … + E(Y6). Using the result that Yi follows a
geometric distribution with success probability (7 – i)/6, we have
6
6
E(Y) = 
= 1 + 6/5 + 6/4 + 6/3 + 6/2 + 6 = 14.7.
i 1 7  i
5.89
Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) =
 y y
1
y1
2
p( y1 , y2 ) – [2(1/3)]2 = 2/9 – 4/9 = –2/9.
y2
As the value of Y1 increases, the value of Y2 tends to decrease.
5.90
From Ex. 5.3 and 5.21, E(Y1) = 4/3 and E(Y2) = 1. Thus,
18
24
 2(1) 12
E(Y1Y2) = 1(1) 84
84  1( 2) 84  1
So, Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) = 1 – (4/3)(1) = –1/3.
5.91
From Ex. 5.76, E(Y1) = E(Y2) = 2/3. E(Y1Y2) =
1 1
 4y
2
1
y22 dy1dy2 = 4/9. So,
0 0
Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) = 4/9 – 4/9 = 0 as expected since Y1 and Y2 are
independent.
1 y2
5.92
From Ex. 5.77, E(Y1) = 1/4 and E(Y2) = 1/2. E(Y1Y2) =
 6y
1
y2 (1  y2 )dy1dy2 = 3/20.
0 0
So, Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) = 3/20 – 1/8 = 1/40 as expected since Y1 and Y2 are
dependent.
5.93
a. From Ex. 5.55 and 5.79, E(Y1Y2) = 0 and E(Y1) = 0. So,
Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) = 0 – 0E(Y2) = 0.
b. Y1 and Y2 are dependent.
c. Since Cov(Y1, Y2) = 0, ρ = 0.
d. If Cov(Y1, Y2) = 0, Y1 and Y2 are not necessarily independent.
5.94
a. Cov(U1, U2) = E[(Y1 + Y2)(Y1 – Y2)] – E(Y1 + Y2)E(Y1 – Y2)
= E(Y12) – E(Y22) – [E(Y1)]2 – [E(Y2)]2
= ( 12  12 ) – (  22   22 ) – ( 12   22 ) = 12   22 .
12   22
b. Since V(U1) = V(U2) =    (Y1 and Y2 are uncorrelated),   2
.
1   22
2
1
2
2
c. If 12   22 , U1 and U2 are uncorrelated.
108
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.95
Note that the marginal distributions for Y1 and Y2 are
y1
–1 0
1
y2
0
1
p1(y1) 1/3 1/3 1/3
p2(y2) 2/3 1/3
So, Y1 and Y2 not independent since p(–1, 0) ≠ p1(–1)p2(0). However, E(Y1) = 0 and
E(Y1Y2) = (–1)(0)1/3 + (0)(1)(1/3) + (1)(0)(1/3) = 0, so Cov(Y1, Y2) = 0.
5.96
a. Cov(Y1, Y2) = E[(Y1 – μ1)(Y2 – μ2)] = E[(Y2 – μ2)(Y1 – μ1)] = Cov(Y2, Y1).
b. Cov(Y1, Y1) = E[(Y1 – μ1)(Y1 – μ1)] = E[(Y1 – μ1)2] = V(Y1).
5.97
a. From Ex. 5.96, Cov(Y1, Y1) = V(Y1) = 2.
b. If Cov(Y1, Y2) = 7, ρ = 7/4 > 1, impossible.
c. With ρ = 1, Cov(Y1, Y2) = 1(4) = 4 (a perfect positive linear association).
d. With ρ = –1, Cov(Y1, Y2) = –1(4) = –4 (a perfect negative linear association).
5.98
Since ρ2 ≤ 1, we have that –1 ≤ ρ ≤ 1 or –1 ≤
5.99
Since E(c) = c, Cov(c, Y) = E[(c – c)(Y – μ)] = 0.
Cov(Y1 ,Y2 )
≤ 1.
V (Y1 ) V (Y2 )
5.100 a. E(Y1) = E(Z) = 0, E(Y2) = E(Z2) = 1.
b. E(Y1Y2) = E(Z3) = 0 (odd moments are 0).
c. Cov(Y1, Y1) = E(Z3) – E(Z)E(Z2) = 0.
d. P(Y2 > 1 | Y1 > 1) = P(Z2 > 1 | Z > 1) = 1 ≠ P(Z2 > 1). Thus, Y1 and Y2 are dependent.
5.101 a. Cov(Y1, Y2) = E(Y1Y2) – E(Y1)E(Y2) = 1 – α/4 – (1)(1) = 

4
.
b. This is clear from part a.
c. We showed previously that Y1 and Y2 are independent only if α = 0. If ρ = 0, if must be
true that α = 0.
5.102 The quantity 3Y1 + 5Y2 = dollar amount spend per week. Thus:
E(3Y1 + 5Y2) = 3(40) + 5(65) = 445.
E(3Y1 + 5Y2) = 9V(Y1) + 25V(Y2) = 9(4) + 25(8) = 236.
5.103 E(3Y1 + 4Y2 – 6Y3) = 3E(Y1) + 4E(Y2) – 6E(Y3) = 3(2) + 4(–1) – 6(–4) = –22,
V(3Y1 + 4Y2 – 6Y3) = 9V(Y1) + 16V(Y2) + 36E(Y3) + 24Cov(Y1, Y2) – 36Cov(Y1, Y3) –
48Cov(Y2, Y3) = 9(4) + 16(6) + 36(8) + 24(1) – 36(–1) – 48(0) = 480.
5.104 a. Let X = Y1 + Y2. Then, the probability distribution for X is
x
1
2
3
p(x) 7/84 42/84 35/84
Thus, E(X) = 7/3 and V(X) = .3889.
b. E(Y1 + Y2) = E(Y1) + E(Y2) = 4/3 + 1 = 7/3. We have that V(Y1) = 10/18, V(Y2) = 42/84,
and Cov(Y1, Y1) = –1/3, so
Chapter 5: Multivariate Probability Distributions
109
Instructor’s Solutions Manual
V(Y1 + Y2) = V(Y1) + V(Y2) + 2Cov(Y2, Y3) = 10/18 + 42/84 – 2/3 = 7/18 = .3889.
5.105 Since Y1 and Y2 are independent, V(Y1 + Y2) = V(Y1) + V(Y1) = 1/18 + 1/18 = 1/9.
5.106 V(Y1 – 3Y2) = V(Y1) + 9V(Y2) – 6Cov(Y1, Y2) = 3/80 + 9(1/20) – 6(1/40) = 27/80 = .3375.
1 1 y2
5.107 Since E(Y1) = E(Y2) = 1/3, V(Y1) = V(Y2) = 1/18 and E(Y1Y2) =
  2 y y dy dy
1
0
2
1
2
= 1/12,
0
we have that Cov(Y1, Y1) = 1/12 – 1/9 = –1/36. Therefore,
E(Y1 + Y2) = 1/3 + 1/3 = 2/3 and V(Y1 + Y2) = 1/18 + 1/18 + 2(–1/36) = 1/18.
5.108 From Ex. 5.33, Y1 has a gamma distribution with α = 2 and β = 1, and Y2 has an
exponential distribution with β = 1. Thus, E(Y1 + Y2) = 2(1) + 1 = 3. Also, since
 y1
E(Y1Y2) =
 y y e
1
2
 y1
dy2 dy1  3 , Cov(Y1, Y1) = 3 – 2(1) = 1,
0 0
V(Y1 – Y2) = 2(1)2 + 12 – 2(1) = 1.
Since a value of 4 minutes is four three standard deviations above the mean of 1 minute,
this is not likely.
5.109 We have E(Y1) = E(Y2) = 7/12. Intermediate calculations give V(Y1) = V(Y2) = 11/144.
1 1
Thus, E(Y1Y2) =
  y y (y
1
2
1
 y2 )dy1dy2  1 / 3 , Cov(Y1, Y1) = 1/3 – (7/12)2 = –1/144.
0 0
From Ex. 5.80, E(30Y1 + 25Y2) = 32.08, so
V(30Y1 + 25Y2) = 900V(Y1) + 625V(Y2) + 2(30)(25) Cov(Y1, Y1) = 106.08.
The standard deviation of 30Y1 + 25Y2 is 106.08 = 10.30. Using Tchebysheff’s
theorem with k = 2, the interval is (11.48, 52.68).
5.110 a. V(1 + 2Y1) = 4V(Y1), V(3 + 4Y2) = 16V(Y2), and Cov(1 + 2Y1, 3 + 4Y2) = 8Cov(Y1, Y2).
8Cov(Y1 ,Y2 )
So,
   .2 .
4V (Y1 ) 16V (Y2 )
b. V(1 + 2Y1) = 4V(Y1), V(3 – 4Y2) = 16V(Y2), and Cov(1 + 2Y1, 3 – 4Y2) = –8Cov(Y1, Y2).
- 8Cov(Y1 ,Y2 )
   .2 .
So,
4V (Y1 ) 16V (Y2 )
c. V(1 – 2Y1) = 4V(Y1), V(3 – 4Y2) = 16V(Y2), and Cov(1 – 2Y1, 3 – 4Y2) = 8Cov(Y1, Y2).
8Cov(Y1 ,Y2 )
   .2 .
So,
4V (Y1 ) 16V (Y2 )
110
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.111 a. V(a + bY1) = b2V(Y1), V(c + dY2) = d2V(Y2), and Cov(a + bY1, c + dY2) = bdCov(Y1, Y2).
bdCov(Y1 ,Y2 )
bd

Y1 ,Y2 . Provided that the constants b and d are
So, W1 ,W2 
b 2V (Y1 ) d 2V (Y2 ) | bd |
nonzero,
bd
is either 1 or –1. Thus, | W1 ,W2 |  | Y1 ,Y2 | .
| bd |
b. Yes, the answers agree.
5.112 In Ex. 5.61, it was showed that Y1 and Y2 are independent. In addition, Y1 has a gamma
distribution with α = 2 and β = 2, and Y2 has an exponential distribution with β = 2. So,
with C = 50 + 2Y1 + 4Y2, it is clear that
E(C) = 50 + 2E(Y1) + 4E(Y2) = 50 + (2)(4) + (4)(2) = 66
V(C) = 4V(Y1) + 16V(Y2) = 4(2)(4) + 16(4) = 96.
5.113 The net daily gain is given by the random variable G = X – Y. Thus, given the
distributions for X and Y in the problem,
E(G) = E(X) – E(Y) = 50 – (4)(2) = 42
V(G) = V(G) + V(G) = 32 + 4(22) = 25.
The value $70 is (70 – 42)/5 = 7.2 standard deviations above the mean, an unlikely value.
5.114 Observe that Y1 has a gamma distribution with α = 4 and β = 1 and Y2 has an exponential
distribution with β = 2. Thus, with U = Y1 – Y2,
a. E(U) = 4(1) – 2 = 2
b. V(U) = 4(12) + 22 = 8
c. The value 0 has a z–score of (0 – 2)/ 8 = –.707, or it is –.707 standard deviations
below the mean. This is not extreme so it is likely the profit drops below 0.
5.115 Following Ex. 5.88:
a. Note that for non–negative integers a and b and i ≠ j,
P(Yi = a, Yj = b) = P(Yj = b | Yi = a)P(Yi = a)
But, P(Yj = b | Yi = a) = P(Yj = b) since the trials (i.e. die tosses) are independent ––
the experiments that generate Yi and Yj represent independent experiments via the
memoryless property. So, Yi and Yj are independent and thus Cov(Yi. Yj) = 0.
b. V(Y) = V(Y1) + … + V(Y6) = 0 +
1/ 6
( 5 / 6 )2
 ( 42//66)2  ( 33//66)2  ( 24//66)2  (15//66)2 = 38.99.
c. From Ex. 5.88, E(Y) = 14.7. Using Tchebysheff’s theorem with k = 2, the interval is
14.7 ± 2 38.99 or (0 , 27.188)
Chapter 5: Multivariate Probability Distributions
111
Instructor’s Solutions Manual
5.116 V(Y1 + Y2) = V(Y1) + V(Y2) + 2Cov(Y1, Y2), V(Y1 – Y2) = V(Y1) + V(Y2) – 2Cov(Y1, Y2).
When Y1 and Y2 are independent, Cov(Y1, Y2) = 0 so the quantities are the same.
5.117 Refer to Example 5.29 in the text. The situation here is analogous to drawing n balls
from an urn containing N balls, r1 of which are red, r2 of which are black, and N – r1 – r2
are neither red nor black. Using the argument given there, we can deduce that:
E(Y1) = np1
V(Y1) = np1(1 – p1)  NN 1n 
where p1 = r1/N
E(Y2) = np2
V(Y2) = np2(1 – p2)  NN 1n 
where p2 = r2/N
Now, define new random variables for i = 1, 2, …, n:
1 if alligator i is a mature female
1 if alligator i is a mature male
Ui  
Vi  
otherwise
otherwise
0
0
n
n
i 1
i 1
Then, Y1  U i and Y2  Vi . Now, we must find Cov(Y1, Y2). Note that:
n
 n

E(Y1Y2) = E  U i , Vi  =
i 1
 i 1

n
 E (U V ) +  E (U V
i 1
i i
i
i j
j
).
Now, since for all i, E(Ui, Vi) = P(Ui = 1, Vi = 1) = 0 (an alligator can’t be both female
and male), we have that E(Ui, Vi) = 0 for all i. Now, for i ≠ j,
E(Ui, Vj) = P(Ui = 1, Vi = 1) = P(Ui = 1)P(Vi = 1|Ui = 1) =
Since there are n(n – 1) terms in
 E (U V
i j
i
j
r1
N
 
r2
N 1
N
N 1
) , we have that E(Y1Y2) = n(n – 1) NN1 p1 p2 .
Thus,
Cov(Y1, Y2) = n(n – 1) NN1 p1 p2 – (np1)(np2) =  n (NN1n ) p1 p2 .
So,
E Yn1  Yn2  1n np1  np2  = p1  p2 ,

V

Y1
n
p1 p2 .


 Yn2 
1
n2
V (Y1 ) V (Y2 )  2Cov(Y1 ,Y2 ) =
N n
n ( N 1)
p
1
 p2  ( p1  p2 ) 2 
5.118 Let Y = X1 + X2, the total sustained load on the footing.
a. Since X1 and X2 have gamma distributions and are independent, we have that
E(Y) = 50(2) + 20(2) = 140
V(Y) = 50(22) + 20(22) = 280.
b. Consider Tchebysheff’s theorem with k = 4: the corresponding interval is
140 + 4 280 or (73.07, 206.93).
So, we can say that the sustained load will exceed 206.93 kips with probability less
than 1/16.
112
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.119 a. Using the multinomial distribution with p1 = p2 = p3 = 1/3,
6
P(Y1 = 3, Y2 = 1, Y3 = 2) = 3!16!!2! 13  = .0823.
b. E(Y1) = n/3, V(Y1) = n(1/3)(2/3) = 2n/9.
c. Cov(Y2, Y3) = –n(1/3)(1/3) = –n/9.
d. E(Y2 – Y3) = n/3 – n/3 = 0, V(Y2 – Y3) = V(Y2) + V(Y3) – 2Cov(Y2, Y3) = 2n/3.
5.120 E(C) = E(Y1) + 3E(Y2) = np1 + 3np2.
V(C) = V(Y1) + 9V(Y2) + 6Cov(Y1, Y2) = np1q1 + 9np2q2 – 6np1p2.
5.121 If N is large, the multinomial distribution is appropriate:
a. P(Y1 = 2, Y2 = 1) = 2!15!!2! (.3) 2 (.1)1 (.6) 2  .0972 .

V
b. E


Y1
n
 Yn2  = p1  p2 = .3 – .1 = .2
Y1
n

Y2
n
1
n2
V (Y1 ) V (Y2 )  2Cov(Y1 ,Y2 ) =
p1q1
n

p2 q2
n
 2 p1np2 = .072.
5.122 Let Y1 = # of mice weighing between 80 and 100 grams, and let Y2 = # weighing over 100
grams. Thus, with X having a normal distribution with μ = 100 g. and σ = 20 g.,
p1 = P(80 ≤ X ≤ 100) = P(–1 ≤ Z ≤ 0) = .3413
p2 = P(X > 100) = P(Z > 0) = .5
a. P(Y1 = 2, Y2 = 1) = 2!41!!1! (.3413) 2 (.5)1 (.1587)1  .1109 .
b. P(Y2 = 4) =
4!
0!4!0!
(.5) 4  .0625 .
5.123 Let Y1 = # of family home fires, Y2 = # of apartment fires, and Y3 = # of fires in other
types. Thus, (Y1, Y2, Y3) is multinomial with n = 4, p1 = .73, p2 = .2 and p3 = .07. Thus,
P(Y1 = 2, Y2 = 1, Y3 = 1) = 6(.73)2(.2)(.07) = .08953.
5.124 Define C = total cost = 20,000Y1 + 10,000Y2 + 2000Y3
a. E(C) = 20,000E(Y1) + 10,000E(Y2) + 2000E(Y3)
= 20,000(2.92) + 10,000(.8) + 2000(.28) = 66,960.
b. V(C) = (20,000)2V(Y1) + (10,000)2V(Y2) + (2000)2V(Y3) + covariance terms
= (20,000)2(4)(.73)(.27) + (10,000)2(4)(.8)(.2) + (2000)2(4)(.07)(.93)
+ 2[20,000(10,000)(–4)(.73)(.2) + 20,000(2000)(–4)(.73)(.07) +
10,000(2000)(–4)(.2)(.07)] = 380,401,600 – 252,192,000 = 128,209,600.
5.125 Let Y1 = # of planes with no wine cracks, Y2 = # of planes with detectable wing cracks,
and Y3 = # of planes with critical wing cracks. Therefore, (Y1, Y2, Y3) is multinomial with
n = 5, p1 = .7, p2 = .25 and p3 = .05.
a. P(Y1 = 2, Y2 = 2, Y3 = 1) = 30(.7)2(.25)2(.05) = .046.
b. The distribution of Y3 is binomial with n = 5, p3 = .05, so
Chapter 5: Multivariate Probability Distributions
113
Instructor’s Solutions Manual
P(Y3 ≥ 1) = 1 – P(Y3 = 0) = 1 – (.95)5 = .2262.
5.126 Using formulas for means, variances, and covariances for the multinomial:
E(Y1) = 10(.1) = 1
V(Y1) = 10(.1)(.9) = .9
E(Y2) = 10(.05) = .5
V(Y2) = 10(.05)(.95) = .475
Cov(Y1, Y2) = –10(.1)(.05) = –.05
So,
E(Y1 + 3Y2) = 1 + 3(.5) = 2.5
V(Y1 + 3Y2) = .9 + 9(.475) + 6(–.05) = 4.875.
5.127 Y is binomial with n = 10, p = .10 + .05 = .15.
10 
a. P(Y = 2) =  (.15) 2 (.85) 8 = .2759.
2
b. P(Y ≥ 1) = 1 – P(Y = 0) = 1 – (.85)10 = .8031.
5.128 The marginal distribution for Y1 is found by

 f (y , y
f1 ( y1 ) 
1
2
)dy2 .

Making the change of variables u = (y1 – μ1)/σ1 and v = (y2 – μ2)/σ2 yields



1
1
f1 ( y1 ) 
exp 
(u 2  v 2  2uv ) dv .
2

21 1   2   2(1   )

To evaluate this, note that u 2  v 2  2uv  ( v  u ) 2  u 2 (1   2 ) so that


( v  u ) 2  dv ,
)
21 1  


So, the integral is that of a normal density with mean ρu and variance 1 – ρ2. Therefore,
1 ( y1 1 )2 / 2 12
f1 ( y1 ) 
e
, –∞ < y1 < ∞,
21
which is a normal density with mean μ1 and standard deviation σ1. A similar procedure
will show that the marginal distribution of Y2 is normal with mean μ2 and standard
deviation σ2.
f1 ( y1 ) 
1
2
e
u 2 / 2

1
 exp  2(1  
2
5.129 The result follows from Ex. 5.128 and defining f ( y1 | y2 )  f ( y1 , y2 ) / f 2 ( y2 ) , which
yields a density function of a normal distribution with mean 1  (1 / 2 )( y2   2 ) and
variance 12 (1  2 ) .
n
n
n
n
5.130 a. Cov(U 1 ,U 2 )   ai b j Cov(Yi ,Y j )  ai b jV (Yi )   2  ai b j , since the Yi’s are
i 1 j 1
i 1
i 1
n
independent. If Cov(U1 ,U 2 ) = 0, it must be true that
a b
i 1
i
j
= 0 since σ2 > 0. But, it is
n
trivial to see if
a b
i 1
i
j
= 0, Cov(U1 ,U 2 ) = 0. So, U1 and U2 are orthogonal.
114
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
b. Given in the problem, (U1 ,U 2 ) has a bivariate normal distribution. Note that
n
n
n
i 1
i 1
i 1
n
E(U1) =   a i , E(U2) =   bi , V(U1) =  2  ai , and V(U2) =  2  bi . If they are
2
2
i 1
orthogonal, Cov(U1 ,U 2 ) = 0 and then U1 ,U 2  0. So, they are also independent.
5.131 a. The joint distribution of Y1 and Y2 is simply the product of the marginals f1 ( y1 ) and
f 2 ( y2 ) since they are independent. It is trivial to show that this product of density has
the form of the bivariate normal density with ρ = 0.
n
b. Following the result of Ex. 5.130, let a1 = a2 = b1 = 1 and b2 = –1. Thus,
a b
i 1
i
j
=0
so U1 and U2 are independent.
5.132 Following Ex. 5.130 and 5.131, U1 is normal with mean μ1 + μ2 and variance 2σ2 and U2
is normal with mean μ1 – μ2 and variance 2σ2.
5.133 From Ex. 5.27, f ( y1 | y2 )  1/ y2 , 0 ≤ y1 ≤ y2 and f 2 ( y2 )  6 y2 (1  y2 ) , 0 ≤ y2 ≤ 1.
a. To find E(Y1 | Y2  y2 ) , note that the conditional distribution of Y1 given Y2 is uniform
y
on the interval (0, y2). So, E(Y1 | Y2  y2 ) = 2 .
2
b. To find E( E(Y1 | Y2 )) , note that the marginal distribution is beta with α = 2 and β = 2.
So, from part a, E( E(Y1 | Y2 )) = E(Y2/2) = 1/4. This is the same answer as in Ex. 5.77.
5.134 The z–score is (6 – 1.25)/ 1.5625 = 3.8, so the value 6 is 3.8 standard deviations above
the mean. This is not likely.
5.135 Refer to Ex. 5.41:
a. Since Y is binomial, E(Y|p) = 3p. Now p has a uniform distribution on (0, 1), thus
E(Y) = E[E(Y|p)] = E(3p) = 3(1/2) = 3/2.
b. Following part a, V(Y|p) = 3p(1 – p). Therefore,
V(p) = E[3p(1 – p)] + V(3p) = 3E(p – p2) + 9V(p)
= 3E(p) – 3[V(p) + (E(p))2] + 9V(p) = 1.25
5.136 a. For a given value of λ, Y has a Poisson distribution. Thus, E(Y | λ) = λ. Since the
marginal distribution of λ is exponential with mean 1, E(Y) = E[E(Y | λ)] = E(λ) = 1.
b. From part a, E(Y | λ) = λ and so V(Y | λ) = λ. So, V(Y) = E[V(Y | λ)] + E[V(Y | λ)] = 2
c. The value 9 is (9 – 1)/ 2 = 5.657 standard deviations above the mean (unlikely score).
5.137 Refer to Ex. 5.38: E(Y2 | Y1  y1 ) = y1/2. For y1 = 3/4, E(Y2 | Y1  3 / 4) = 3/8.
5.138 If Y = # of bacteria per cubic centimeter,
a. E(Y) = E(Y) = E[E(Y | λ)] = E(λ) = αβ.
Chapter 5: Multivariate Probability Distributions
115
Instructor’s Solutions Manual
b. V(Y) = E[V(Y | λ)] + V[E(Y | λ)] = αβ + αβ2 = αβ(1+β). Thus,   (1  ) .
 n  n
5.139 a. E (T | N  n )  E  Yi    E (Yi )  n .
 i 1  i 1
b. E (T )  E[ E (T | N )]  E ( N)   . Note that this is E(N)E(Y).
5.140 Note that V(Y1) = E[V(Y1 | Y2)] + V[E(Y1 | Y2)], so E[V(Y1 | Y2)] = V(Y1) – V[E(Y1 | Y2)].
Thus, E[V(Y1 | Y2)] ≤ V(Y1).
5.141 E(Y2) = E( E(Y2 | Y1 )) = E(Y1/2) =

2
2 2
V(Y2) = E[V(Y2 | Y1)] + V[E(Y2 | Y1)] = E[ Y / 12 ] + V[Y1/2] = (2λ )/12 + (λ )/2 =
.
3
2
1
5.142 a. E(Y) = E[E(Y|p)] = E(np) = nE(p) =
2
2
n
.
 
b. V(Y) = E[V(Y | p)] + V[E(Y | p)] = E[np(1 – p)] + V(np) = nE(p – p2) + n2V(p). Now:
n
n(  1)
nE(p – p2) =
–
   (  )(     1)
n2V(p) =
So, V(Y) =
n 2 
.
(  ) 2 (    1)
n
n(  1)
n(    n)
n 2 
–
+
=
.
2
   (  )(     1)
(  ) (    1) (  ) 2 (    1)
5.143 Consider the random variable y1Y2 for the fixed value of Y1. It is clear that y1Y2 has a
normal distribution with mean 0 and variance y12 and the mgf for this random variable is
m(t )  E(ety1Y2 )  et
2 2
y1
/2
.

Thus, mU (t )  E (e )  E (e
tU
tY1Y2
)  E[ E (e
tY1Y2
| Y1 )]  E (e
tY12 / 2
)


1
2
e  y1 / 2 (1t ) dy1 .
2
2
Note that this integral is essentially that of a normal density with mean 0 and variance
1
, so the necessary constant that makes the integral equal to 0 is the reciprocal of the
1t 2
standard deviation. Thus, mU (t )  1  t 2  . Direct calculations give mU (0)  0 and
mU (0)  1 . To compare, note that E(U) = E(Y1Y2) = E(Y1)E(Y2) = 0 and V(U) = E(U2) =
E(Y12Y22) = E(Y12)E(Y22) = (1)(1) = 1.
1 / 2
116
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.144 E[ g (Y1 )h(Y2 )]   g ( y1 )h( y2 ) p( y1 , y2 )  g ( y1 )h( y2 ) p1 ( y1 ) p2 ( y2 ) 
y1
y2
 g( y ) p ( y ) h( y
1
1
1
y1
y1
2
y2
) p2 ( y2 ) E[ g (Y1 )]  E[ h(Y2 )] .
y2
5.145 The probability of interest is P(Y1 + Y2 < 30), where Y1 is uniform on the interval (0, 15)
and Y2 is uniform on the interval (20, 30). Thus, we have
30 30 y 2
 1  1 
P(Y1  Y2  30)      dy1dy2 = 1/3.
15  10 
20 0 
5.146 Let (Y1, Y2) represent the coordinates of the landing point of the bomb. Since the radius
is one mile, we have that 0 ≤ y12  y 22 ≤ 1. Now,
P(target is destroyed) = P(bomb destroys everything within 1/2 of landing point)
This is given by P(Y12  Y22  ( 12 ) 2 ) . Since (Y1, Y2) are uniformly distributed over the unit
circle, the probability in question is simply the area of a circle with radius 1/2 divided by
the area of the unit circle, or simply 1/4.
5.147 Let Y1 = arrival time for 1st friend, 0 ≤ y1 ≤ 1, Y2 = arrival time for 2nd friend, 0 ≤ y2 ≤ 1.
Thus f (y1, y2) = 1. If friend 2 arrives 1/6 hour (10 minutes) before or after friend 1, they
will meet. We can represent this event as |Y1 – Y2| < 1/3. To find the probability of this
event, we must find:
P(| Y1  Y2 |  1 / 3) 
5.148 a. p( y1 , y2 ) 
1 / 6 y1 1 / 6
2
 4  3 

    

 y1   y2   3 y1  y2 
9
 
 3

1dy2 dy1 
0
0
5 / 6 y1 1 / 6

1dy2 dy1 
1 / 6 y1 1 / 6
1
1
 1dy dy
2
1
 11 / 36 .
5 / 6 y1 1 / 6
, y1 = 0, 1, 2, 3, y2 = 0, 1, 2, 3, y1 + y2 ≤ 3.
b. Y1 is hypergeometric w/ r = 4, N = 9, n = 3; Y2 is hypergeometric w/ r = 3, N = 9, n = 3
c. P(Y1 = 1 | Y2 ≥ 1) = [p(1, 1) + p(1, 2)]/[1 – p2(0)] = 9/16
y1
1
5.149 a. f1 ( y1 )   3 y1dy2  3 y , 0 ≤ y1 ≤ 1, f1 ( y1 )   3 y1 dy1  23 (1  y 22 ) , 0 ≤ y2 ≤ 1.
2
1
0
y2
b. P(Y1  3 / 4 | Y2  1/ 2)  23 / 44 .
c. f(y1 | y2) = 2 y1 /(1  y 22 ) , y2 ≤ y1 ≤ 1.
d. P(Y1  3 / 4 | Y2  1 / 2)  5 / 12 .
5.150 a. Note that f(y2 | y1) = f(y1, y2)/f(y1) = 1/y1, 0 ≤ y2 ≤ y1. This is the same conditional
density as seen in Ex. 5.38 and Ex. 5.137. So, E(Y2 | Y1 = y1) = y1/2.
Chapter 5: Multivariate Probability Distributions
117
Instructor’s Solutions Manual
1
b. E(Y2) = E[E(Y2 | Y1)] = E(Y1/2) =

y1
2
3 y12 dy1 = 3/8.
0
1
c. E(Y2) =
y
3
2 2
(1  y22 )dy2 = 3/8.
0
5.151 a. The joint density is the product of the marginals: f ( y1 , y 2 )  12 e  ( y1  y2 ) /  , y1 ≥ ∞, y2 ≥ ∞
a a  y2
1
2
0 0
b. P(Y1  Y2  a )  

e ( y1  y2 ) /  dy1dy2 = 1 – 1  a / e  a /  .
5.152 The joint density of (Y1, Y2) is f ( y1 , y2 )  18( y1  y12 ) y22 , 0 ≤ y1 ≤ 1, 0 ≤ y2 ≤ 1. Thus,
1
P(Y1Y2 ≤ .5) = P(Y1 ≤ .5/Y2) = 1 – P(Y1 > .5/Y2) = 1 –
1
  18( y
1
 y12 ) y 22 dy1dy 2 . Using
.5 .5 / y 2
straightforward integration, this is equal to (5 – 3ln2)/4 = .73014.
5.153 This is similar to Ex. 5.139:
a. Let N = # of eggs laid by the insect and Y = # of eggs that hatch. Given N = n, Y has a
binomial distribution with n trials and success probability p. Thus, E(Y | N = n) = np.
Since N follows as Poisson with parameter λ, E(Y) = E[E(Y | N )] = E(Np ) = λp.
b. V(Y) = E[V(Y | N)] + V[E(Y | N)] = E[Np(1 – p)] + V[Np] = λp.
5.154 The conditional distribution of Y given p is binomial with parameter p, and note that the
marginal distribution of p is beta with α = 3 and β = 2.
1
1
 n  1 y 2
a. Note that f ( y )   f ( y, p )  f ( y | p ) f ( p )dp  12  p (1  p ) n y 1 dp . This
 y0
0
0
integral can be evaluated by relating it to a beta density w/ α = y + 3, β = n + y + 2.
Thus,
 n  ( n  y  2)( y  3)
f ( y )  12 
, y = 0, 1, 2, …, n.
( n  5)
 y
b. For n = 2, E(Y | p) = 2p. Thus, E(Y) = E[E(Y|p)] = E(2p) = 2E(p) = 2(3/5) = 6/5.
5.155 a. It is easy to show that
Cov(W1, W2) = Cov(Y1 + Y2, Y1 + Y3)
= Cov(Y1, Y1) + Cov(Y1, Y3) + Cov(Y2, Y1) + Cov(Y2, Y3)
= Cov(Y1, Y1) = V(Y1) = 2ν1.
b. It follows from part a above (i.e. the variance is positive).
118
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
5.156 a. Since E(Z) = E(W) = 0, Cov(Z, W) = E(ZW) = E(Z2 Y 1 / 2 ) = E(Z2)E( Y 1 / 2 ) = E( Y 1 / 2 ).
This expectation can be found by using the result Ex. 4.112 with a = –1/2. So,
( 2  12 )
Cov(Z, W) = E( Y 1 / 2 ) =
, provided ν > 1.
2 ( 2 )
b. Similar to part a, Cov(Y, W) = E(YW) = E( Y W) = E( Y )E(W) = 0.
c. This is clear from parts (a) and (b) above.
 
y 
( y  ) 1
y 1e [( 1) / ]
5.157 p( y )   p( y |  ) f ()d  
, y = 0, 1, 2, … . Since
d 
( y  1)() 
( y  1)() 
0
0
it was assumed that α was an integer, this can be written as



 y    1    1 

p( y )  
 
 , y = 0, 1, 2, … .
y    1     1 

y
5.158 Note that for each Xi, E(Xi) = p and V(Xi) = pq. Then, E(Y) = ΣE(Xi) = np and V(Y) = npq.
The second result follows from the fact that the Xi are independent so therefore all
covariance expressions are 0.
5.159 For each Wi, E(Wi) = 1/p and V(Wi) = q/p2. Then, E(Y) = ΣE(Xi) = r/p and V(Y) = rq/p2.
The second result follows from the fact that the Wi are independent so therefore all
covariance expressions are 0.
5.160 The marginal probabilities can be written directly:
P(X1 = 1) = P(select ball 1 or 2) = .5
P(X2 = 1) = P(select ball 1 or 3) = .5
P(X3 = 1) = P(select ball 1 or 4) = .5
P(X1 = 0) = .5
P(X2 = 0) = .5
P(X3 = 0) = .5
Now, for i ≠ j, Xi and Xj are clearly pairwise independent since, for example,
P(X1 = 1, X2 = 1) = P(select ball 1) = .25 = P(X1 = 1)P(X2 = 1)
P(X1 = 0, X2 = 1) = P(select ball 3) = .25 = P(X1 = 0)P(X2 = 1)
However, X1, X2, and X3 are not mutually independent since
P(X1 = 1, X2 = 1, X3 = 1) = P(select ball 1) = .25 ≠ P(X1 = 1)P(X2 = 1)P(X1 = 3).
Chapter 5: Multivariate Probability Distributions
119
Instructor’s Solutions Manual
5.161 E (Y  X )  E (Y )  E ( X )  1n  E (Yi )  m1  E ( X i )  1   2
V (Y  X )  V (Y ) V ( X )  n12 V (Yi )  m12 V ( X i )  12 / n   22 / m
5.162 Using the result from Ex. 5.65, choose two different values for α with –1 ≤ α ≤ 1.
5.163 a. The distribution functions with the exponential distribution are:
F1 ( y1 )  1  e  y1 , y1 ≥ 0;
F2 ( y2 )  1  e y2 , y2 ≥ 0.
Then, the joint distribution function is
F ( y1 , y2 )  [1  e  y1 ][1  e  y2 ][1  (e  y1 )( e  y2 )] .
Finally, show that
2
F ( y1 , y 2 ) gives the joint density function seen in Ex. 5.162.
y1y 2
b. The distribution functions with the uniform distribution on (0, 1) are:
F1 ( y1 ) = y1, 0 ≤ y1 ≤ 1 ;
F2 ( y2 ) = y2, 0 ≤ y2 ≤ 1.
Then, the joint distribution function is
F ( y1 , y2 )  y1 y2 [1  (1  y1 )(1  y2 )] .
2
F ( y1 , y 2 ) = f ( y1 , y 2 )  1  (1  2 y1 )(1  2 y 2 ), 0 ≤ y1 ≤ 1, 0 ≤ y2 ≤ 1.
c.
y1y 2
d. Choose two different values for α with –1 ≤ α ≤ 1.


5.164 a. If t1 = t2 = t3 = t, then m(t, t, t) = E e t ( X1  X 2  X 3 ) . This, by definition, is the mgf for the
random variable X1 + X2 + X3.


b. Similarly with t1 = t2 = t and t3 = 0, m(t, t, 0) = E e t ( X1  X 2 ) .
c. We prove the continuous case here (the discrete case is similar). Let (X1, X2, X3) be
continuous random variables with joint density function f ( x1 , x2 , x3 ) . Then,

m(t1 , t2 , t3 ) 


  e
t1 x1 t2 x2
e
e t3 x3 f ( x1 , x2 , x3 )dx1dx2 dx3 .
  
Then,

 k1  k 2 k3
m(t1 , t2 , t3 ) t1 t2 t3 0  
t1k1 t2k2 t3k3



 x
k1
1
x2k2 x3k3 f ( x1 , x2 , x3 )dx1dx2 dx3 .
 
This is easily recognized as E X 1k1 X 2k2 X 3k3 .
5.165 a. m(t1 , t2 , t3 )   x1!xn2!!x3!et1x1 t2 x2 t3 x3 p1x1 p2x2 p3x3
x1
=
x2
x3

x1
x2
n!
x1!x2 !x3 !
( p1et1 ) x1 ( p2 et2 ) x2 ( p3et3 ) x3 = ( p1e t1  p2 e t2  p3 e t3 ) n . The
x3
final form follows from the multinomial theorem.
120
Chapter 5: Multivariate Probability Distributions
Instructor’s Solutions Manual
b. The mgf for X1 can be found by evaluating m(t, 0, 0). Note that q = p2 + p3 = 1 – p1.
c. Since Cov(X1, X2) = E(X1X2) – E(X1)E(X2) and E(X1) = np1 and E(X2) = np2 since X1 and
X2 have marginal binomial distributions. To find E(X1X2), note that
2
m(t1 , t2 ,0) t1 t2 0  n( n  1) p1 p2 .
t1 t2
Thus, Cov(X1, X2) = n(n – 1)p1p2 – (np1)(np2) = –np1p2.
5.166 The joint probability mass function of (Y1, Y2, Y3) is given by
 N 1  N 2  N 3   Np1  Np2  Np3 

    


y1  y 2  y 3   y1  y 2  y 3 

p( y1 , y 2 , y 3 ) 

,
N 
N
 
 
n
n
where y1 + y2 + y3 = n. The marginal distribution of Y1 is hypergeometric with r = Np1, so
E(Y1) = np1, V(Y1) = np1(1–p1)  NN 1n  . Similarly, E(Y2) = np2, V(Y2) = np2(1–p2)  NN 1n  . It
can be shown that (using mathematical expectation and straightforward albeit messy
algebra) E(Y1Y2) = n( n  1) p1 p2 NN1 . Using this, it is seen that
Cov(Y1, Y2) = n( n  1) p1 p2 NN1 – (np1)(np2) = –np1p2  NN 1n  .
(Note the similar expressions in Ex. 5.165.) Finally, it can be found that
p1 p2
.

(1  p1 )(1  p2 )
5.167 a. For this exercise, the quadratic form of interest is
At 2  Bt  C  E (Y12 )t 2  [ 2 E (Y1Y2 )]t  [ E (Y22 )]2 .
Since E[(tY1 – Y2)2] ≥ 0 (it is the integral of a non–negative quantity), so we must have
that At 2  Bt  C ≥ 0. In order to satisfy this inequality, the two roots of this quadratic
must either be imaginary or equal. In terms of the discriminant, we have that
B 2  4 AC  0 , or
[ 2 E (Y1Y2 )]2  4 E (Y12 ) E (Y22 )  0 .
Thus, [ E (Y1Y2 )]2  E (Y12 ) E (Y22 ) .
b. Let μ1 = E(Y1), μ2 = E(Y2), and define Z1 = Y1 – μ1, Z2 = Y2 – μ2. Then,
2 
by the result in part a.
[ E (Y1  1 )(Y2   2 )] 2
[ E ( Z1 Z 2 )] 2

1
[ E (Y1  1 ) 2 ]E[(Y2   2 ) 2 ] E ( Z12 ) E ( Z 2 2 )
Download